The input images are coadded to produce a total signal map.
The observations are divided into two groups (with alternate files going into each group) which are coadded separately. These coadds are subtracted from one another to create the jack-knife map.
The azimuthally-average angular power spectrum of the jack-knife map (which should consist purely of noise) is calculated and used to remove residual low-spatial frequency noise from the signal map and the given (map-filtered) psf. This is the so-called whitening step (because it produces a map which has a noise power spectrum that is white).
The whitened signal map is processed with a matched filter using the whitened psf image as the psf.
The jack-knife map is also whitened and processed with the matched filter. This map should consist purely of noise.
Signal-to-noise ratio maps are created for the filtered versions of the signal map and the jack-knife map.
The outcome (the match-filtered whitened signal map with suffix _mf) should be the optimal map with white noise properties. This is the map to be used for science goals.
Ideally there should be an even number of observations, but this is not important if the number of input files is large.
A fuller description of the procedure may be found in the documentation for the SCUBA-2 recipe REDUCE_SCAN_FAINT_POINT_SOURCES_JACKKNIFE.
It is recommended that the PICARD recipe SCUBA2_JACKKNIFE_PSF be run to create a suitable PSF to use for this recipe.